ES2360898T3 - CODING BY TRANSFORMED, USING WEIGHING WINDOWS AND WITH SMALL DELAY. - Google Patents

Abstract

Decoding method, by transform, of a signal represented by a succession of frames that have been encoded using at least two types of weighting windows, of respective respective lengths, where, in case of receiving a passing information from a window long to a short window and to compensate for a direct transition from the long window to the short window: - (63) samples (b) are determined from a decoding by applying a type of short synthesis window (61) to a given frame (T'i + 1) that was encoded using a short analysis window and - complementary samples are obtained (67, 69): - partially decoding (DCT -1) a frame (T'i) that precedes the given frame and which was coded using a type of long analysis window and - applying a combination of at least two values: - from samples, respectively, of the given frame (T'i + 1) and at least one frame (T'i ) that precedes the plot da da (T'i + 1), - and weighted by respective weighting functions (w1, n, w2, n; w'1, n, w'2, n) tabulated and stored in memory of a decoder.

Description

The present invention relates to the decoding of audio digital signals.

In a transformed coding scheme, for a reduction of the information flow, it is usually sought to reduce the agreed accuracy for the coding of the samples, ensuring that the listener can only, however, perceive as little as possible of degradations.

To this end, the precision reduction, carried out by a quantification operation, is controlled from the psychoacoustic model. This model, based on a knowledge of the properties of the human ear, allows the quantification noise to be dosed at the less noticeable auditory frequencies.

In order to use the information derived from the psychoacoustic model, essentially data in the frequency domain, it is classic to perform a time / frequency transformation, quantifying in this frequency domain.

Figure 1 illustrates, schematically, the structure of an encoder per transform, with:

-a bank of analysis filters FA1, ... FAn, which affects the input signal X,

-a quantization module Q, followed by a COD coding module,

-and a BS bank of synthesis filters FS1, ..., FSn, which provides the encoded signal X ’.

To also reduce the flow rate before transmission, quantified frequency samples are coded, often with the help of a coding called "entropic" (lossless coding). The quantification is done, in a classical way, by a scalar quantifier, uniform or not, or also by a vector quantifier.

The noise introduced in the quantification stage is formed by the synthesis filter bank (also called "reverse transformation"). The inverse transformation linked to the analysis transform must, therefore, be chosen in order to adequately concentrate the quantization noise, frequently or temporarily, in order to avoid making it audible.

The analysis transformation should concentrate, as best as possible, the signal energy in order to allow easy coding of the samples in the transformed domain. In particular, the encoding gain of the transform, which depends on the input signal, should be maximized as much as possible. For this purpose, a relationship of the type is used:

SNR = GTC + K - R

Where K is a constant term that can, advantageously, assume a value of 6.02.

Thus, the signal to noise ratio obtained (SNR) is to provide the number of bits per selected sample (R) increased in the Gτc component that represents the encoding gain of the transform. The more important the coding gain, the better the quality of the reconstruction.

The importance of the transformation in coding is therefore understood. It guarantees an easy coding of the samples, thanks to its aptitude to concentrate, at the same time, the energy of the signal (by the part of analysis and the noise of quantification by the part of synthesis).

Since the audio signals are notoriously non-stationary, it is appropriate to adapt the time / frequency transformation over time, depending on the nature of the audio signal.

Some applications to usual coding techniques are described below.

In the case of modulated transforms, standardized audio coding techniques integrate filter banks, modulated in the trigonometric function of the cosine, which allow these coding techniques to be implemented with the help of fast algorithms, based on transforms in the cosine function or fast Fourier transform.

Among the transforms of this type, the most frequently used transform (in MP3, MPEG-2 and MPEG-4 AAC encoding, in particular) is the MDCT transform ("Modified discrete cosine transform") whose expression is presented below:

image 1

with the following notations:

 M represents the magnitude of the transform  χn + tM are the samples of the digitized sound with a period 1 (inverse of the sampling frequency) in

Faith

the instant n + tM,  t is the frame rate.

It is a basic function of the transform, whose h (n) is called a prototype filter of magnitude 2M.

To restore the initial temporary samples, the following inverse transformation is applied in order to reconstitute the samples 0 <n <M - 1:

image2

image 1

Referring to Figure 1a, the reconstruction is performed as follows:

t

 reverse DCT transform (referred to below as DCT-1) of the Xk samples that provide 2M samples,

t1

 Inverse DCT transform of the Xk samples giving 2M samples, the first M samples presenting a temporary support identical to the last M samples of the previous frame.

 Weighting by the synthesis window h (M + n) for the second half of the Ti frame (M last samples) and by the synthesis window h (n) for the first half of the next frame Ti + 1 (M first samples) and

 Additions of the components in windows on the common support.

In order to ensure the exact reconstruction (called “perfect”) of the signal (according to the condition xn + tM = χn + tM), it is convenient to choose a prototype window h (n) that satisfies some limitations.

Under typical conditions, the following relationships are satisfactory in order to allow a perfect reconstruction:

image 1

the windows presenting an even symmetry with respect to a central sample.

It is relatively simple to satisfy these two simple limitations and, in this respect, a classic prototype filter is constituted by a sine window that is written as follows:

image 1

Of course, there are other forms of prototype filter, such as the windows defined in the MPEG-4 standard under the name “Kaiser Bessel Dérivées Derivatives” (or KBD) or also the weak overlap windows (Low Overlap).

An example of treatment by an MDCT transform, with long windows, is shown in Figure 1a. In this Figure:

-the arrows in dashed lines illustrate a subtraction,

-the arrows in solid lines illustrate an addition,

-the arrows in mixed dashed lines illustrate a DCT treatment for coding and DCT-1 for DEC decoding, this term DCT corresponding to the cosine term of the base function presented above,

-the samples of the signal to be encoded are in a flow called xin and the evolution of the coding / decoding treatments of some samples marked and referred to as a and b in Figure 1b and as e and f in Figure 1c is followed,

-xin samples are regrouped by frames, a current frame being indicated as Ti, while the preceding and following frames are indicated, respectively by Ti-1 and Ti + 1,

-The DEC reference is typical of the treatment performed by the decoder (using FS synthesis windows with addition - overlap),

-the analysis windows are indicated as FA and the synthesis windows are indicated as FS,

-n is the distance between the central part of the window and the sample a.

The reference calc T'i is proper for the calculation of the encoded frame T'i using the FA analysis window and the respective samples of the Ti-1 and Ti frames. It is simply, in this case, an example of an agreement illustrated in Figure 1a. It could also be decided, for example, to index the Ti and Ti + 1 frames for the calculation of a T'i encoded frame. Following the example of Figure 1a, the reference calc T'i + 1 is typical of the calculation of the T'i + 1 encoded frame using the respective samples of the Ti and Ti + 1 frames.

The terms v1 and v2 obtained before the DCT transformation and the inverse DCT-1 transformation are obtained with equations of the type:

v1 = a * h (M + n) + b * h (2 * M-1-n), and

v2 = b * h (M-1-n) - a * h (n)

Thus, after the overall DCT / DCT-1 treatment and the synthesis window, the reconstruction terms a 'and b' are written as follows:

a '= v1 * h (M + n) - v2 * h (n) = a * h (M + n) * h (M + n) + b * h (2 * M-1-n) * h (M + n) - b * h (M-1-n) * h (n ) + a * h (n) * h (n),

Y

b '= v1 * h (2 * M-1-n) + v2 * h (M-1-n) = a * h (M + n) * h (2M-n-1) + b * h (2 * M-1-n-1) * h (2M-n-1) + b * h (M-1- n) * h (M-1-n) - a * h (n) * h (M-1-n) and it is verified, then, that the reconstruction is perfect (a '= a and b' = b). (using the relations (1) and deducing from them h (M-1-n) = h (n + M))

The principle described above of an MDCT transform naturally extends to so-called ELT ("Extended Lapped Transform") transforms, where the order of the base functions is greater than twice the magnitude of the transform with, in particular:

image 1 and L = 2KM, where K is a positive integer greater than 2.

For reconstruction, instead of associating two consecutive frames as for an MDCT transform, the synthesis of the samples implements successive frames in windows.

It is further indicated that the restriction of symmetry of the windows (principle described in detail below) can be relaxed for an ELT transform. The identity restriction between the analysis and synthesis windows can also be relaxed and, then, we speak of bi-orthogonal filters.

In view of the need to adapt the transform to the signal to be encoded, prior techniques, such as the one proposed by Omar Niamut et al in “RD Optimal Time Segmentation for the Time - Varying MOCT”, EURASSIP; they authorize what is called a "window switching", that is, a change in the course of time of the magnitude of the transform used.

The need to change window length can be justified, in particular, in the following case. When the signal to be encoded, for example, a word signal, presents a transient (non-stationary) signal that characterizes a strong attack (for example, the pronunciation of a “ta” or “pa” sound that characterizes an occlusive syllable in the word signal), it is advisable to increase the temporal resolution of the coding and, furthermore, decrease the magnitude of the coding windows, which needs, therefore, to move from a long window to a short window. Very precisely, in the known prior art, in this case, it is passed from a long window (Figure 2a which will be described later) to a transition window (Figure 2c described below) and then, to a series of short windows (Figure 2b described below). It is necessary, therefore, to anticipate an attack on at least one next frame, as will be seen in detail below, before being able to decide the length of the coding window of a current frame and consequently, encode the current frame.

An example of changing window length according to the prior art is described below.

A typical example is the change in magnitude of an MDCT transform of magnitude M to a magnitude M / 8, as provided in the MPEG-AAC standard.

In order to retain the perfect reconstruction property, equation (1) above must be replaced by the following formulas, at the time of transition between two quantities:

image 1

A relationship is also provided for consecutive prototype filters of different magnitudes:

image 1

There is, therefore, a symmetry around the magnitude M / 2 at the time of transition.

Different types of window have been illustrated in Figures 2a to 2e, with respectively: - a sinusoidal window (sinusoidal symmetric function) of magnitude 2M = 512 samples for Figure 2a, - a sinusoidal window (sinusoidal symmetric function) of magnitude 2Ms = 64 samples for Figure 2b, - a transition window that allows to pass from a magnitude 512 to 64 for Figure 2c, - a transition window that allows to pass from a magnitude of 64 to 512 for Figure 2d, - and an example of construction made with the help of the base windows presented above,

for Figure 2e.

In each sequence it is of a predetermined "length" that defines what is called "the window length". In this way, samples to be encoded are combined, at least two to two, and weighted, in combination, by respective window weighting values, as seen with reference to Figure 1a.

More particularly, the sinusoidal windows (Figures 2a and 2b) are symmetrical, that is, that the weighting values are practically equal, to one and another part, of a central value, in the middle of the sequence of values that form the window . A preferred embodiment consists in choosing "sine" functions to define the variations of weighting values of these windows. Other window choices are still possible (for example, those used in MPEG AAC encoders).

On the contrary, it will be deduced that the transition windows (Figures 2c and 2d) are asymmetrical and have a “flat” zone (PLA reference), which means that the weighting values, in these areas, are maximum and for example , are equal to "1". As will be seen with reference to Figures 1b and 1c, using a transition window from a long window to a short window (Figure 2c), two samples (in the example shown in Figure 1b), of which the sample a is simply weighted by a factor "1", while the sample b is weighted by the factor "0", in the calculation of the T'i encoded frame, so that the two samples, of which the samples a are simply transmitted , such as the DCT set apart in the T'i encoded frame.

The use of a variable magnitude transform in a coding system is described below. It also describes the operations at the level of a decoder that must reconstitute the audio samples.

In standardized systems, the encoder selects, very often, the transform to be used over time. Thus, in the AAC standard, the encoder transmits two bits that allow you to select one of the four window magnitude configurations presented above.

The MDCT transform treatment, which uses the transition windows (long - short) is illustrated in Figures 1b and 1c. These figures represent the calculations performed, in the same way as for Figure 1a.

In Figures 1b and 1c, only a few short analysis windows have been shown, with FA references (with Ms = M / 2 in the illustrated example). However, in reality, as illustrated in Figure 2e, a succession of several short windows (with Ms = M / 8) is expected. It will be understood, therefore, that each FA window of Figures 1b and 1c actually encompasses a succession of short windows.

The FTA transition window (Figure 1b), for the calculation of the T'i encoded frame (reference calc T’i) presents:

-a long semi-window in M samples, on its rising edge and

-in its lowering flank:

or a first flat zone PLA (with weighting values equal to 1) in (M / 2-Ms / 2) samples,

or a short half-window of descent in Ms samples and

or a second flat zone (with weighting values equal to 0) in (M / 2-Ms / 2) samples.

For the calculation of the following encoded frame T'i + 1 (reference calc T'i + 1), the first (M / 2-Ms / 2) samples are ignored and therefore not treated by short windows, being the following Ms samples weighted by the rising edge of the short FA analysis window, as illustrated in Figures 1b and 1c and the following Ms samples are weighted by their falling edge.

In the following, the following notations are used:

-M is the magnitude of long frame,

-Ms is the short plot magnitude.

In Figure 1b, sample b is synthesized using only the short windows to respect the analogy with the calculation for the long windows. Next, due to the particular shape of the long-short transition half-window, the sample a is reconstructed directly from the analysis and synthesis transition windows. The transition window is an FTA reference object in Figures 1b and 1c.

In Figure 1c, the samples corresponding to the transition zone between the long-short window and the short window are calculated. By analogy with the calculation for the long windows of Figure 1a, the treatment of the samples named e and f (marked) is followed here.

Two examples of window transition situation are described below.

In a first example, an attack was detected that requires the use of short windows in audio signal 5 at an instant t = 720 (Figure 2e). The encoder must indicate to the decoder the use of long-short transition windows that interpose between the long windows previously used and the short rear windows. Thus, the encoder indicates to the decoder, successively, the sequences:  long window  long-short transition window  short window 15  short-long transition window  long window. The decoder then applies a relationship of the type:

image 1

where

image 1 Y

image 1 they represent the functions of synthesis of the transformed ones in the times t and t + 1, that can be different from each other.

The reconstruction is carried out as before, with the difference that if the basic functions

image 1 Y

image 1 They are

Different "quantities" are then made, in reference to Figure 1b:

 an inverse DCT transform of magnitude M of the samples

image 1 which provide 2M samples,

 an inverse DCT transform of magnitude Ms of the samples

image 1 giving 2 Ms samples, the first Ms presenting a common temporary support of length Ms in an overlapping area comprising the ascending part of the short window, with the samples derived from the DCT transform

Inverse of magnitude M of the descending part of the FTA transition window,

 a multiplication by the dual synthesis window of the FTA transition window and with FTS reference in Figure 1b, for the first half and a multiplication by the short synthesis window FS for the second half and

 the additions of these components in the window over the overlapping zone, the temporary support corresponding to an end part of the initial frame Ti.

The decoder is therefore slave of the encoder and faithfully applies the window types decided by the encoder.

In this first example, the encoder detects a transition at the time of the arrival of the samples of a first frame (for example, frame 1 of Figure 2e, which contains the samples between the instants t = 512 and t = 767). The encoder can then decide that the current window should be a long-short transition window encoded, transmitted and signaled to the decoder. Then, eight short windows are applied successively between the samples t = 624 and t = 911. Thus, at the time of the transition (t = 720), the encoder uses the short windows, which allows a better temporal representation of the signal.

In a second example, a transition to the sample t = 540 is detected. When the encoder receives the samples from

A first frame (frame 0 of Figure 2e for example), does not detect the transition and therefore selects a long window. At the time of the arrival of the samples of a second frame (frame 1 in the example of Figure 2e), next, the encoder detects an attack (at time t = 540). In this case then, the detection is carried out too late and the use of a transition window does not allow to benefit from the use of short temporary supports (short windows) at the time of the attack. The encoder must then anticipate the use of short windows and therefore insert an additional coding delay corresponding to at least M / 2 samples.

It will be understood, then, that a drawback of the known prior art lies in the fact that it is necessary to introduce a supplementary delay in the encoder to allow an attack to be detected in the time signal of a next frame and then anticipate the passage to the short windows This "attack" can correspond to a transient signal of great intensity, such as an occlusive syllable, for example, in a word signal or also the appearance of a percussive signal in a musical sequence.

In some telecommunications applications, the additional delay required for the detection of transient signals and the use of transition windows is not permissible. Thus, for example, in the MPEG-4 AAC "Low Delay" encoder, short windows are not used and only long windows are allowed.

The present invention aims to improve this operational situation.

It discloses a transition between windows that does not require the introduction of a supplementary delay.

Considers, in this regard, a method of decoding, by transform, a decoder by transform and a computer program, intended to be stored in memory of an encoder by transform, of an audiodigital signal represented by a sequence of frames,

where:

- at least two weighting windows, of different respective lengths, are expected and

-a short window is used to encode a frame where a particular event has been detected.

This particular event can be, for example, a non-stationary phenomenon, such as a strong attack present in the audiodigital signal that contains the current plot.

More particularly, for the coding of a current frame, it is sought to detect the particular event in this current frame and:

- at least if the particular event is detected at the beginning of the current frame, a short window is applied to encode the current frame,

-While if the particular event is not detected in the current frame, a long window is applied to encode the current frame.

These steps are repeated for a following frame, so that it is possible, within the meaning of the invention, to encode a given frame using a long window and encode a frame that immediately follows this given frame directly using, then, a short window , without using a transition window as in the prior art.

Thus, it is possible to move directly from a long window to a short window, and the detection of the particular event can be carried directly on the frame in the process of coding and no longer in the next frame, as in the prior art. From this moment on, a coding carried out by the method in the sense of the invention is carried out, without additional delay, with respect to a MDCT transform of fixed magnitude, contrary to the prior art codings.

Other features and advantages of the invention will appear upon examining the following detailed description and the accompanying drawings where, in addition to Figures 1, 1a, 1b, 1c, 2a, 2b, 2c, 2d, 2e, related to the prior art and described previously:

- Figure 3a illustrates, in a schematic representation, a coding / decoding treatment according to an embodiment of the invention, following the evolution of samples a and b, as in Figure 1b described above,

- Figure 3b illustrates, in a schematic representation, a coding / decoding treatment according to an embodiment of the invention, following the evolution of samples e and f, as in Figure 1c described above and

- Figures 4a and 4b illustrate an example of variation of the weighting functions used for decoding compensation, performed in the implementation of the invention, 8

-Figure 5a illustrates an example of treatment that can be applied in an encoder according to the invention,

- Figure 5b illustrates an example of treatment that can be applied in a decoder according to the invention and

-Figure 6 illustrates the respective structures of an encoder and a decoder and the communication of the window type information used in the coding,

-Figure 7 illustrates a long synthesis window for the case of an ELT transform with M = 512 components and an overlap coefficient K = 4,

- Figure 8 represents the appearance of the weighting functions w1, n and w2, n (for n between 0 and M / 2-Ms / 2) in an embodiment in which the influence of the samples is taken into account passed in an overlapping coding context,

-Figure 9 represents the appearance of the weighting functions w'1, n and w'2, n (for n between M / 2-Ms / 2 and M / 2 + Ms / 2) in this embodiment,

-Figure 10 represents the appearance of the weighting functions w'3, n and w'4, n (for n between M / 2-Ms / 2 and M / 2 + Ms / 2) in this embodiment,

- Figure 11 represents the appearance of the weighting functions w1, n and w2, n in the entire range of n between 0 and M / 2 + Ms / 2 in a variant of the embodiment illustrated in Figure 8,

-Figure 12 represents the appearance of the weighting functions w3, n and w4, n in the entire range of n between 0 and M / 2 + Ms / 2 in this variant.

The present invention makes it possible to avoid applying transition windows at least for the passage from a long window to a short window.

Thus, taking up the second example described above, with reference to Figure 2e, if a non-stationary phenomenon or "attack" is detected at time t = 540, the present invention proposes to use a long window for frame 0 (window to be extends from instant t = 256 to instant t = 511). Then, at the time of the acquisition of the samples of the following frame (t = 512 at = 767) and the detection of an attack at t = 540, the encoder uses eight short windows to encode the samples from the moment t = 368 (corresponding at t = 512-M / 2-MS / 2), so far t = 655 (corresponding at t = 512 + M / 2 + MS / 2-1, where

- 2 * M = 512 is the magnitude of the long window and

- 2 * MS = 64 is the magnitude of the short window, in the example described),

and this, without the classic, asymmetric transition window as shown in Figures 1b and 1c relative to the prior art.

At the decoder level, at the time of receiving the encoded frame with short windows, the decoder then proceeds to the following operations:

 reception of information derived from the encoder indicating that short windows should be used for the current frame,

 application of an advantageous treatment to compensate for the direct transition from a long window to a short window in the coding, being an example of this treatment described in more detail below, referring to Figure 5b.

Figures 3a and 3b illustrate the method of decoding according to the invention to obtain, on one hand, the samples a and b that are in an area without overlapping between the long and short windows (Figure 3a) and on the other hand, the samples e and f that they are in this overlap zone (Figure 3b). In particular, this overlapping zone is defined by the lowering edge of the long window FL and the rising edge of the first short window FC.

Thus, with reference to Figures 3a and 3b, for coding, the samples of frames Ti-1 and Ti are weighted by the long analysis window FL to constitute the encoded frame T'i and the samples of the next frame Ti and Ti + 1 are weighted directly by short FC analysis windows, without applying the transition window.

It will also be noted with reference to Figures 3a and 3b, that the first short analysis window FC is preceded by values not taken into account by the short windows (for the samples preceding the sample e, in the example of the Figure 3b) More particularly, this treatment is applied to the first M / 2-Ms / 2 samples of the frame to be encoded T'i + 1, and this, analogously to the prior art encoders / decoders. In general, it is sought to disturb as little as possible the treatments performed in the coding, as well as in the decoding, with respect to the prior art. For this reason it is decided, for example, to ignore the first samples of the T'i + 1 encoded frame.

Of course, in Figures 3a and 3b, only the case of two short windows (Ms = M / 2) of FC analysis has been represented. However, it is envisaged, as in the prior art, a succession of several short windows and each succession of short windows is illustrated in these Figures 3a and 3b bearing the reference FC.

In the following, two embodiments are described for decoding a frame T'i + 1, which has been encoded using a short window FC, while an immediately preceding frame T'i, had been encoded using a long window FL.

In a first embodiment, it is completely freed from the use of synthesis windows in decoding and it is shown that the perfect reconstruction property is ensured.

In Figure 3a, at the time of detecting an attack that needs a window change (from a long window directly to a short window), the samples are first synthesized from the short windows only (sample b in Figure 3a). Then, the effect of sample b calculated above is compensated in the value v1 calculated from the long analysis window. The coding calculation (T'i encoded frame) for sample a is then carried out as follows:

v1 = a * h (M + n) + b * h (2 * M-1-n).

On the contrary, the sample a is not weighted in the coding value v2, since the calculation of weighting, from the short window, followed by a combination is carried out on a different temporary support (encoded frame Ti + 1) and It is had after reconstruction from the short windows:

v2 = b

The perfect reconstruction in the coding / decoding according to the invention is preferably verified. Indeed:

a '= (v1 - v2 * h (2 * M-1-n)) / h (M + n) = a

It will also be noted, in the decoding, that the samples taken from the values v2 = b and the following should be determined first, before the determination of the samples of the beginning of the frame (as sample a). Therefore, a temporary investment is made at the time of decoding.

In Figure 3b, the coded samples of the transition zone between the long window FL (descending front flank) and the first short window FC (rising front flank) are therefore calculated at the level, therefore, of the samples e and f. The expression of the coded coefficients (or "following v1 and v2 values") is provided, in this overlapping zone between the two FL and FC windows, by the following equations:

v1 = e * h (M + n) + f * h (2 * M-1-n) and

v2 = f * hS (MS-1-m) - e * hS (m)

In the decoder, this system of equations with two unknowns must therefore be solved to find the values of the samples e and f:

e = [v 1 * hs (Ms- 1 -m) -v2 * h (2 * M- 1 -n)] / [h (M + n) * hs (Ms- 1 -m) + hs (m) * h (2 * M- 1 -n)]

f = [v1 * hs (m) + v2 * h (M + n)] / [hs (Ms-1-m) * h (M + n) + h (2 * M-1-n) * hs ( m)]

The formulas that advantageously verify the property of the perfect reconstruction are also deduced from the foregoing.

e '= [v1 * hs (Ms + m) - v2 * h (n)] / [h (M + n) * hs (Ms + m) + h (2 * M-1-n) * hs (m )] = e,

Y

f = [v1 * hs (2 * Ms-1-m) + v2 * h (M-1-n)] / [h (M + n) * hs (Ms + m) + h (2 * M-1 -n) * hs (m)] = f,

with m = n - M / 2 + Ms / 2

It will be deduced that the value v2 is weighted by the long window h and, contrary to what was established in the prior art (where v2 was weighted by the short window hs as indicated in the lower part of Figure 1c).

In a second embodiment, synthesis windows are preserved for decoding. They are in the same way as the analysis windows (homologous or dual of the analysis windows) such as those illustrated in Figures 3a and 3b and carry the FLS reference for a long, synthesis window, and the FCS reference for a window short, of synthesis. This second embodiment has the advantage of being subject to the operation of the decoders of the prior art, namely, using a long synthesis window to decode a frame that has been encoded with a long analysis window and using a series of Short synthesis windows to decode a frame that has been encoded with a series of short analysis windows.

On the contrary, a correction of these synthesis windows is applied, by "compensation", to decode a frame that has been encoded with a long window, while it should have been encoded with a long-short transition window. In other words, to compensate for the effect of direct passage from a long window to a short window, in the encoder, the following treatment is performed to decode a current frame T'i + 1 that has been encoded using a short window FC, while an immediately preceding frame T'i would have been encoded using a long window FL.

The equations given above for decoding and linking samples a, b, e, f to values v1 and v2 can be rewritten in the form of weighted sums of two terms, as follows, in particular by making a temporary inversion.

First, it is placed in the first short FCS synthesis windows and after the overlapping zone mentioned above (usually in the sample v2 = b and the following ones in the illustration, for explanatory purposes of Figure 3a). For decoding this part without overlapping, starting from short FCS synthesis windows only, the "values" of the T'i + 1 encoded frame from v2 = b are decoded first (Figure 3a). After decoding samples b and the following, the weighted sum of the following two terms applies:

image3

encoding / decoding is perfect reconstruction),

~

-notation l designates what would correspond to samples that have been decoded (application of

n

a reverse transformation DCT-1) using a FLS long synthesis window, without correction and -s represents fully decoded samples (usually sample b and the samples that

n

continue) with the help of the succession of short windows of FCS synthesis. The two weighting functions W1, n and W2, n are written, then:

image 1

~

It will be understood that the "samples" l are, in fact, incompletely decoded values by synthesis and

Weighting using the long synthesis window. These are, typically, the values v1 of Figure 3a, multiplied by the coefficients h (M + n) of the FLS window and where also samples from the start of the Ti frame, such as sample a, are also involved.

fifteen

25

35

Four. Five

It will also be noted that the samples b and the following are determined here in the first place and are written in the formula given above "sM-1-n", thus illustrating the temporary inversion proposed by the decoding treatment in this second form of realization.

It is also observed that the weighting made by the long FLS synthesis window is avoided because it is removed from the term W1, n (due to the division by h (M + n)).

On the other hand, for the reconstruction of the covered sample part, at the same time, by the long window FL (descending front flank) and the first short window FC (ascending front flank), which correspond to the area of the eaf samples of Figure 3b usually applies the following combination of the two weighted terms:

image 1

~

As before, the terms 1 constitute the values incompletely reconstituted by synthesis and the

n

~

weighting by the long FLS synthesis window and the terms s represent the values incompletely

n

reconstituted from the rising front flank of the first short synthesis window FCS. The weighting functions w'1, n and w'2, n are given here by:

image 1

All these weighting functions W1, n, w2, n, w'n1, n and w'2, n consist of por frozen ’elements that only depend on long and short windows. Examples of variation of such weighting functions are presented in Figures 4a and 4b. In an advantageous embodiment, the values taken by these functions can be calculated a priori (tabulated) and definitely stored in a decoder according to the invention.

Thus, with reference to Figure 5b, the decoding treatment of a T'i frame that has been encoded by passing directly from a long analysis window to a short analysis window, can comprise the following steps, in an exemplary embodiment . To decode the T'i frame (step 60), a short synthesis window (step 61) is first applied to decode the value v2 = b (step 63) at the end of the frame. It is based, in this case, on a coded frame following T'i + 1 (step 62) to determine b. A long synthesis window (step 64) is then applied to decode the samples at the start of T'i frame (step 65), applying the compensation image 1

for any number between 0 and M / 2-Ms / 2 with the help of the relationship

 (step 67) and using the pre-calculated and tabulated weighting values w1, n and w2, n (step 66).

The decoding of the "central" zone of the T'i encoded frame (between e and f), and therefore for n between M / 2-Ms / 2 and M / 2 + Ms / 2, can be carried in parallel ( sign "+" of Figure 5b) using, at the same time, the short and long synthesis windows (step 68) and applying, in particular, the compensation (step 69) from the relationship image 1

and with the weighting values w'1, n and w'2, n precalculated and tabulated (step 70). Finally, this treatment (step 71) deduces the values of all types of samples a, b, e or f from the initial frame Ti.

The first and second embodiments described above, for decoding a T'i frame that has been encoded by passing directly from a long analysis window to a short analysis window, guarantee a perfect reconstruction and then allow, for coding, effectively move from a long window to a short window, directly.

It is now described, with reference to Figure 5a, an exemplary embodiment in which it is proposed to free, for coding, the application of a long-short transition window, at least in some cases.

Upon receipt of a Ti frame (step 50), the presence of a non-stationary phenomenon, such as an ATT attack (test 51), is searched in the audio-digital signal directly present in this Ti frame. As long as no phenomenon of this type is detected (arrow n at the exit of test 51), the application of long windows (step 52) is continued for the coding of this Ti frame (step 56). If this is not the case (arrow or at the exit of test 51), it is sought to determine if the ATT phenomenon is at the beginning (for example, in the first half) of the current frame Ti (test 53), in which case (arrow or the test output 53) directly applies a short window (step 54), and more precisely a series of short windows, for the coding of the Ti frame (step 56). This embodiment allows, therefore, to avoid a transition window and not wait for the next Ti + 1 frame to apply a short window.

Thus, it will be understood that, contrary to the state of the art, a particular event such as a non-stationary phenomenon can be detected directly in the current frame of Ti coding and not in a subsequent Ti + 1 frame. The coding delay, according to the invention, is then reduced with respect to that of the prior art. Indeed, if the non-stationary phenomenon is detected at the beginning of the current frame, a short window is applied directly, while in the prior art, it would have been necessary to detect the non-stationary phenomenon in a subsequent Ti + 1 frame so that You can apply a transition window to the current Ti encoding frame.

Referring still to Figure 5a, if the non-stationary phenomenon is detected at the end (for example, in the second half) of the current frame Ti (arrow na at the exit of test 53), a window may preferably be applied Transition (step 55) for coding the current frame Ti (step 56), before applying a succession of short windows. This embodiment allows, in particular, to offer a treatment equivalent to that of the state of the art, guaranteeing a reduction in the coding delay.

Thus, in more generic terms, it is envisaged, in this embodiment, at least three weighting windows:

-

a short window,

-

a long window and

-

a transition window to move from a long window use to a window use

short,

and if a particular event, such as a non-stationary phenomenon, is detected at the end of the current frame (step 53) a transition window (step 55) is applied to encode (step 56) the current frame (Ti).

In a variant of this embodiment, it can be provided, for a passage between a long window use to a short window use:

-for a current Ti frame, use a long FL window,

-and for an immediately consecutive Ti + 1 frame, directly use a short FC window, without using transition window, when even the particular event was detected at the end of the current frame.

This variant has the following advantage. As the encoder must send a window type change information to the decoder, this information can be encoded in a single bit because the decoder no longer has to be informed of the choice between a short window and a transition window.

However, a transition window can be retained to pass from a short window to a long window and in particular, to ensure, in addition, the transmission of the window type change information in a single bit, after receipt of a step information from the long window to the short window, the decoder being able to in this regard:

-use the short window,

- then, in the absence of receiving window type change information, use a transition window from a short window to a long window,

-and then, finally, use a long window.

Figure 6 illustrates the communication of window type information used for encoding, from an encoder 10 to a decoder 20. It is recalled that the encoder 10 comprises a detection module 11 of a particular event, such as a strong attack in the signal contained in a Ti frame in the course of coding and which deduces from this detection the type of window to be used. In this regard, a module 12 selects the type of window to be used and transmits this information to the coding module 13, which delivers the encoded frame T'i using the analysis window FA, selected by module 12. The encoded frame T ' i is transmitted to decoder 20, with the information INF on the type of window used for coding (generally in the same flow). The decoder 20 comprises a synthesis window selection module 22 FS, according to the information INF received from the encoder 10 and the module 23 applies the decoding of the frame T'i to deliver a

decoded frame T.

i

The present invention further considers an encoder such as the encoder 10 of Figure 6 for the implementation of the method according to the invention and more particularly, for the implementation of the treatment illustrated in Figure 5a, or its variant previously described (transmission of window type change information in a single bit).

The present invention also considers a computer program, intended to be memorized in such an encoder and comprising instructions for the implementation of such a treatment or its variant, when such a program is executed by an encoder processor. To this title, Figure 5a can represent the organization chart of such a program.

It is recalled that the encoder 10 uses analysis windows FA and the decoder 20 can use synthesis windows FS, according to the second embodiment mentioned above, these synthesis windows being homologous to the analysis windows FA, proceeding, however, to the compensation correction described above (using the weighting functions W1, n, W2, n, W'1, n and W'2, n).

The present invention further relates to another computer program, intended to be memorized in a decoder by transform, such as the decoder 20 illustrated in Figure 6 and comprising instructions for the implementation of decoding according to the first embodiment or according to the second embodiment described above with reference to Figure 5b, when said program is executed by a processor of this decoder 20. In this title, Figure 5b can represent the flow chart of such a program.

The present invention also relates to the decoder by transform, by itself, which then comprises a memory that stores the instructions of a computer program for decoding.

In generic terms, the method of decoding, according to the invention, by transformed, of a signal represented by a succession of frames that have been encoded using at least two types of weighting windows, of respective respective lengths, is performed as follows.

In case of receiving a passing information from a long window to a short window:

- samples (of type b) are determined from a decoding that applies a type of FCS short synthesis window to a given frame Ti + 1 that has been encoded an FC short analysis window and

- complementary samples are obtained:

 partially decoding (application of a DCT-1 inverse transform) a frame T'i that precedes the given frame and was encoded using a type of long analysis window FL and

 applying a combination of two weighted terms by intervening weighting functions that can be tabulated and stored in the memory of a decoder.

In the second previous embodiment, these are the functions called W1, n, W2, n, W'1, n, W'2, n.

However, this generic decoding treatment is applied in both cases of the first and second embodiments.

In the second embodiment:

- first determine (step 63 of Figure 5b) the samples (b) derived from the weft (T'i + 1) and

- the above is deduced (steps 65–67) samples (a) that temporarily correspond to the beginning of the preceding frame (T'i), these samples being derived from a decoding applying a long FLS synthesis window and own in the second way of realization.

In this case, to:

- a plot presenting M samples,

- a long window presenting 2M samples,

- a short window presenting 2Ms samples, with Ms being less than M, the samples

image 1

, for n between 0 and (M / 2-Ms / 2), n = 0 corresponding to the beginning of a frame in the process of decoding, are given by a combination of two weighted terms of the type:

image4

~

-l are values (v1) derived from the preceding frame T'i,

n

-SM-1-n are samples (b) already decoded using short synthesis windows applied to the given frame T'i + 1 and

-W1, n and W2, n are weighting functions, whose values taken as a function of n can be tabulated and stored in the decoder's memory.

If not, for n between (M / 2-Ms / 2) and (M / 2 + Ms / 2), the samples xn are given by a combination of two weighted terms of the type:

image 1

~

-l are v1 values derived from the preceding frame T'i,

n

~

-S are v2 values derived from the given frame T'i + 1 and

m

-w'1, n and w'2, n are weighting functions, whose values taken as a function of n can also be tabulated and stored in decoder memory.

The present invention allows, therefore, to offer the transition between windows with a reduced delay in relation to the prior art, while retaining the perfect reconstruction property of the transform. This method can be applied with all types of windows (non-symmetric windows and different analysis and synthesis windows) and for different transforms and filter banks.

The compensation treatments presented above, in the case of a transition from a long window to a window of shorter magnitude, extend naturally and analogously to the case of a transition from a short window to a window of greater magnitude. In this case, the absence of a short-long transition window can be compensated in the decoder by a weighting similar to the case presented above.

The invention is then applied to any encoder per transform, in particular those intended for interactive conversational applications, as in the MPEG-4 "AAC-Low Delay" standard, but also to transforms other than MDCT transforms, in particular those transformed before cited ELT ("Extended Lapped Transform") and its bi-orthogonal extensions.

However, in the case of a particular ELT transform, it was observed that the terms of temporary retraction due to modulation (v1) can be combined with terms of temporary retraction from the past. Thus, the corrective treatment presented above takes into account a phenomenon of influence (or “aliasing”) of future samples. On the contrary, the development presented below also takes into account past components in order to cancel them to obtain a perfect reconstruction, at least in the absence of quantification. Therefore, it is proposed to define here a supplementary weighting function that, associated with the synthesized past signal, allows the terms of temporal retraction to be suppressed.

The following is an example of an ELT transform, the one described in the document: "Modulated Filter Banks with Arbitrary System Delay: Efficient Implementations and the Time-Varying Case", Gerald DT Schuller, Tanja Karp, IEEE Transactions on Signal Processing, Vol. 48, No. 3 (March 2000).

The following exemplary embodiment proposes, within the framework of the present invention, a step, without transition, between a long window (for example, 2048 samples) and a short window (for example 128 samples).

* Transformed with long window (K = 4, M = 512)

It is a small delay transform, whose window is magnitude KM = 2048 and whose analysis is written in the form:

image 1

image5

-M being the number of spectral components obtained,

, the notation of the input signal in window and

s

--w nw n being the notation of the long synthesis window.

LD L

Figure 7 illustrates this long synthesis window for the case of an ELT transform with M = 512 components and an overlap coefficient K = 4.

The inverse transform is written:

image 1

and the reconstructed signal xn + tM is obtained by adding four elements overlap (K = 4):

image 1

It should be noted that the synthesis window is defined as follows:

image 1

while the analysis window is defined from the synthesis window by inversion of the order of the samples, that is:

image 1

* Transformed with short window (K = 2, Ms = 64)

The analysis transform is written, in the case of a short window, in the form:

image 1

image5

with:

as a window input signal and -wS (n), as a short synthesis window. The inverse transform is written:

image 1

and the reconstructed signal xn + tM is obtained by adding two elements overlapping (κs = 2):

image 1

In this notation, t is the short plot index and the analysis and synthesis windows are identical, since they are symmetric, with:

image 1

* Expressions of weighting functions

In this embodiment, to:

-

a plot that presents M samples,

-

a long window that presents 4M samples,

-

a short window that presents 2Ms samples, being Ms less than M,

for n between 0 and M / 2-Ms / 2 (n = 0 corresponding to the beginning of a frame in the process of decoding), Jcn samples are given by a combination of four weighted terms of the type:

image 1

~

-x represents the decoded samples (corresponding to the initial samples xn if the

n

encoding / decoding is perfect reconstruction),

designates what would correspond to samples that have been incompletely decoded from the frame (T'i) that precedes the given frame (T'i + 1) (application of an inverse transformation), using a long synthesis window with addition to the memory items image 1

precedents

 without correction of the plot T’i,

-sn represents fully decoded samples with the help of the succession of short windows of FCS synthesis of the T'i + 1 frame (for samples of index n such as M / 2 + MS / 2 <n <M) and samples

-w1, n, w2, n, w3, n and w4, n are weighting functions, whose values taken as a function of n can be tabulated and stored in decoder memory or calculated based on the long analysis and synthesis windows and short

In a preferred embodiment, the following expressions can be chosen as weighting functions, in particular with a view to ensuring perfect reconstruction:

image6

image7

image8

image 1

It will be noted that the forms of w3, n and w2, n are somewhat different from those set forth above in the case of the MDCT transform. In effect, the filters are no longer symmetrical (so that the term h2 disappears) and the modulation terms are changed, which explains the change in sign.

Next, always in this embodiment, for n between M / 2-Ms / 2 and M / 2 + Ms / 2, the samples xˆ are given by a combination of four weighted terms of the type:

n

image9

According to the same notations:

~

-l are the incompletely decoded samples of the T’i frame that precedes the given T’i + 1 frame,

n

~

-s are incompletely decoded samples of the first short window of the given frame T’i + 1 and

m

-s represents fully decoded samples for the previous frames (T’i-1, T’i-2,…) and

n

-w'1, n, w'2, n, w'3, n and w'4, n are weighting functions whose values taken as a function of n can also be tabulated and stored in decoder memory or calculated according to the windows of long and short analysis and synthesis.

In a preferred embodiment, weighting functions can be chosen according to the following forms in order to ensure perfect reconstruction:

image8

image 1

Thus, in this embodiment, at the time of a transition between long window and short window, the signal is reconstructed from the combination: - of a weighted version of the reconstructed samples from the short windows, - of a weighted version of partially reconstructed samples from the long window (which image 1 integrates the terms of memory), -and of a weighted version of a combination of the samples of the last synthesized signal. In a variant of this embodiment, it will be noted that the functions w'3n and w'4, n differ little from one another.

Only the terms h (4M -1 - n) and h (3M + n) differ in their expression. One embodiment may, for example, consist in preparing the terms h (4M -1 -n) Sn-2M + h (3M + n) sM-1-n, and then weighing the result by a function that is expressed by:

image 1

and that corresponds, therefore, to the functions w’3, n and w’4, n from which the contributions of the terms h (4M -1 -n) and h (3M + n). This same principle applies, analogously, to w3, n and w4, n. In another variant, the synthesis memory is weighted. In a preferred embodiment, this weighting

may

be a put to zero of the synthesis reports, so that the incompletely

rebuilt, from the long window, be added to a memory

 weighted In this

In this case, the weighting applied to the past synthesized signal may be different.

Figures 9 and 10 show the characteristic forms of the weighting functions w and w ’obtained in the embodiment described above. In particular, referring to the ordinate values of these graphs, it seems that the functions w'3, n and w'4, n represented in Figure 10, can be neglected (taking into account their values taken) with respect to the functions w'1, n and w'2, n represented in Figure 9. The terms in which the functions w'3, n and w'4, n intervene could therefore be omitted in the sum

image10

~ x

n

 which was previously given, with a view to rebuilding the signal

~ x

n

. This omission would bring with it a

Small reconstruction error.

In a variant aimed at achieving greater simplicity of treatment, it is also found that w’3, n and w’4, n are very similar. Therefore, it could be planned not to use more than a combination of these two weights, for example, an average of the two functions, in order to gain in calculation time.

The comparison of Figures 8 (representing the appearance of the weighting functions w1, n and w2, n) and 12 (representing the appearance of the weighting functions w3, n and w4, n) uses the same observations for the functions w3, n and w4, n with respect to the functions w1, n and w2, n.

Therefore, the preceding expressions of

n

~ x

:, if the weights for functions w3, n and w4, n are omitted,

image11

image12

with for example

image 1 or any other linear combination of these two functions, which would result in a moderate reconstruction error.

It should be noted that the omission of the weights by the functions w3, n and w4, n results in an error of

10 reconstruction of a power of 84 dB under the signal and that the use of a simple linear combination (average of these functions, for example) brings, by itself an error of 96 dB under the signal which, both in one case As in another, it is already very satisfactory for audio applications. It is indicated that a perfect reconstruction currently allows to measure in practice an error power of 120 to 130 dB under the signal.

In addition, by not using the terms sn-2M and s-M-1-n in the weighting [1], it is possible to avoid quantification noise staggering from the past. Therefore, an imperfect reconstruction is exchanged in the absence of quantification against a limitation of quantization noise when the signal is encoded in fine.

It is also observed that, in the temporary support 0-128 (Figures 8 and 12), the weighting functions have the 20 particular forms:

image 1

This observation is explained by the shape of the window h {n) (Figure 7) which presents, in the example described, a first part of zero amplitude between 0 and 128. Consequently, in this example, it is preferable, in terms of complexity, decompose the first reconstruction into two phases:

image13

In an embodiment of advantageous algorithmic structure, the weighting functions w1, n and w2, n (Figure 11), of one part and w3, n and w4, n (Figure 12) of another part, can be defined in the entire range from 0 a (M + Ms) / 2, as set forth below.

In a first stage, a primary expression (called ~ x) of the signal xˆ a is calculated from 0 to (M + Ms) / 2

nn

rebuild, as follows:

image 1 (where the calculation of the function w1, n, illustrated in the entire range of n between 0 and M / 2 + Ms / 2 in Figure 11, as well as functions w3, n and w4, n calculated in this same range e 40 illustrated in Figure 12).

Then, for n between 0 and M / 2-Ms / 2 (n = 0 corresponding to the beginning of a frame in the process of decoding), it is stated:

corresponding to the beginning of the curve with reference w2, n in Figure 11 (before abscissa 224),

and for n between M / 2-Ms / 2 and M / 2 + Ms / 2, it is stated:

image14

image 1

which corresponds to the end of the curve referenced w2, n in Figure 11 (after abscissa 224).

5 This distinction of particular treatment for the weighting of functions w2, n and w’2, n is explained as follows.

For each function w1, n and w3, n and w4, n it is only possible to use a variation between 0 and M / 2 + Ms / 2. On the contrary, for the functions w2, n and w’2, n. 10 -w2 function, n ponder the fully decoded samples,

-While the w’2 function, n weights incompletely decoded samples.

15 On the other hand, a “temporary inversion” of the treatment for the w2, n only (index of s in -n) and not for the w’2, n.

Thus, to summarize, through generic terms, this development that allows reducing the influence of past samples, for complete decoding of the samples, at the time of a transition from a long window

20 (with an overlap K> 2) towards a short window (with an overlap K '<K), the decoded samples are obtained by combining at least two weighted terms that intervene the past synthesis signal.

Claims (11)

1. Method of decoding, by transform, a signal represented by a succession of frames that have been encoded using at least two types of weighting windows, of respective respective lengths, where, in case of receiving a passing information from a long window to a short window and to compensate for a direct transition from the long window to the short window: - (63) samples (b) are determined from a decoding by applying a type of short synthesis window (61) to a given frame (T'i + 1) that was encoded using a short analysis window and - complementary samples are obtained (67, 69):  partially decoding (DCT-1) a frame (T'i) that precedes the given frame and that was encoded using a type of long analysis window and  applying a combination of at least two values: - from samples, respectively, of the given frame (T'i + 1) and at minus one frame (T'i) that precedes the given frame (T'i + 1), - and weighted using respective weighting functions (w1, n, w2, n; w’1, n, w’2, n) tabulated and stored in memory of a decoder.

2.

 Method according to claim 1, characterized in that: - samples (b) derived from the given frame (T'i + 1) are determined first (63) and - (65-67) corresponding samples (a) are deduced therefrom. temporarily at the beginning of the previous frame

(T'i), these samples being the result of a decoding that applies a long synthesis window.

3.

 Method according to claim 2, wherein: - a frame has M samples, - a long window has 2M samples, - a short window has 2M samples, with Ms being less than M

characterized in that the samples xˆ, for n between 0 and (M / 2-Ms / 2) corresponding n = 0 at the beginning of n a frame in the process of decoding, is given by a combination of two weighted terms of the type: characterized in that the samples xˆ, for n between (M / 2-Ms / 2) and (M / 2 + Ms / 2), corresponding n = 0 image 1

-

nl ~ are values (v1) derived from the preceding frame (T'i),

-

they are samples (b) already decoded using short synthesis windows applied to the given frame (T'i + 1) and

-

nw1, and nw2, are weighting functions whose values taken, as a function of n, are tabulated and stored in the decoder's memory.

Four.

Method according to one of claims 1 to 3, wherein:

-

a plot presents M samples,

-

a long window presents 2M samples,

-

a short window presents 2Ms samples, being Ms less than M,

n at the beginning of a frame in the process of decoding, they are given by a combination of two weighted terms of the type: image2 ~ -l are values (v1) derived from the previous frame (T'i), n ~ -s are values (v2) derived from the given frame (T'i + 1) and m -w'1, n and w'2, n are weighting functions whose values taken, as a function of n, are tabulated and stored in the decoder memory. 5. Method according to one of claims 1 to 4, characterized in that, for decoding frames encoded by overlapping transformed coding, with a view to reducing the influence of past samples, the signal to be decoded is reconstructed from a combination between:

-

a weighting of reconstructed samples from short windows,

-

a weighting of partially reconstructed samples from a long window and

-

a sample weighting of the last decoded signal.

6.

Method according to claim 5, characterized in that, with:

-

a plot that presents M samples,

-

a long window that presents 4M samples,

-

a short window that presents 2Ms samples, being Ms less than M

for a sample index n between 0 and M / 2-Ms / 2, n = 0 corresponding to the beginning of an ongoing frame ~ decoding, the samples x to be decoded are given by a combination of four weighted terms n of the kind: image2 image3  designates incompletely decoded samples of the frame (T'i) that precedes the given frame (T'i + 1) using a long synthesis window with addition, without correction, to previous memory elements called t index a frame index, -s represents fully decoded samples with the help of a succession of short windows n of synthesis (FCS) of the given frame (T'i + 1), for M / 2 + Ms / 2 <n <M and fully decoded samples of previous frames (T'i, T'i-1, T'i -2,…) for -2M <n <M and -w, w, w and wson, respectively, first, second, third and fourth functions of 1, n 2, n 3, n 4, n weighting dependent on the sample index n and the values taken at least by the first and second weighting functions w and w, depending on n, are tabulated and stored in the 1, n 2, n decoder memory. 7. Method according to one of claims 5 and 6, characterized in that, with: image4

-

a plot that presents M samples,

5

- a long window that presents 4M samples, a short window that presents 2Ms samples, being Ms lower than M,

10

for n between M / 2-Ms / 2 and M / 2 + Ms / 2, a combination of four weighted terms of the type: the samples nxˆ  to  decode come given by a

image5 ~ -l are incompletely decoded samples of the frame (T'i) that precedes the given frame (T'i + 1), n ~ 15 -s are incompletely decoded samples of the first short window of the given frame (T'i + 1), with m m = n –M / 2 + Ms / 2, -s represents fully decoded samples of previous frames (T'i, T'i-1, T'i-2, ...), n 20 -w'1, n, w'2, n, w'3, n and w'4, n are respectively, first, second, third and fourth weighting functions dependent on n and the values taken at least by the first and second Weighting functions w'1, n and w'2, n, depending on n, are tabulated and stored in the decoder's memory. 25 8. Fourth weighting functions are neglected in the calculation of samples xˆ, so that only the values taken by the first and second weighting functions n image6 depending on n, they are tabulated and stored in the decoder's memory. Method according to one of claims 6 and 7, characterized in that the third and fourth functions of 10. Method according to claims 6 and 7, characterized in that: ~~ - it is calculated for n, which varies from 0 to (M + Ms) / 2, a primary expression x of the signal x to be decoded, nn 40 according to a weighted combination of type: image7 image8 image6 -for n between 0 and M / 2-Ms / 2, corresponding n = 0 at the beginning of a frame in the process of decoding, it is stated: image9 -for n between M / 2-Ms / 2 and M / 2 + Ms / 2 it is stated: image10 11. Transformer decoder, of a signal represented by a succession of frames derived from an encoder that use at least two types of weighting windows, of respective respective lengths, characterized in that it has at least: - means of receiving a step information from a long window to a short window and, to compensate for a direct transition from the long window to the short window: - means for determining samples (b) from a decoding that applies a type of short synthesis window (61) to a given frame (T'i + 1) that was encoded using a short analysis window, as well as: - means of obtaining complementary samples (67, 69) that are capable of:  partially decode (DCT-1) a frame (T'i) that precedes the given frame and that was encoded using a type of long analysis window and  apply a combination of at least two values: -sample of samples, respectively, of the given frame (T'i + 1) and of at least one frame (T'i) that precedes the given frame (T'i + 1), -and weighted by respective weighting functions tabulated and stored in the memory of a decoder. image6

12.

Computer program, intended to be stored in memory of a decoder by transform, characterized in that it comprises instructions for the implementation of the decoding method according to one of claims 1 to 10, when executed by a processor of such a decoder.

13.

Transformer decoder according to claim 11, characterized in that it comprises a memory that stores the instructions of a computer program according to claim 12.